Research on Colorization Algorithm for γ-Photon Flow Field Images Using the PECN Model

Abstract γ-photon tomography technology, which leverages the high penetration ability of high-energy γ-photons and their electrical neutrality, has become a groundbreaking non-contact method for monitoring industrial flow fields. However, γ-photon flow fields appear as grayscale images. To enable parameter inversion of these fields, the grayscale images must be colorized into flow field parameter density maps, such as temperature and pressure maps. The DeOldify algorithm is commonly used for coloring grayscale images. γ-photon flow field images exhibit probabilistic imaging features, leading to "color gradient jumps" when DeOldify is applied. The PECN (Physical Enhancement Colorization Network) model forms the basis of the γ-photon flow field image colorization algorithm proposed in this paper to address this challenge. This study introduces a U-Net+GAN framework that integrates multi-scale attention and physical perception augmentation, using ResNet101 as the generator backbone. To enhance physical consistency and reduce color gradient jumps, the discriminator adaptively fuses global and local information across channels and spatial dimensions and incorporates improved blocks to better distinguish boundaries and textures. A flow field condition monitoring experiment was performed on the SW60B turbojet engine at 50% throttle using the Gate simulation platform to evaluate the algorithm’s performance. The validation focused on colorizing the random selected γ-photon flow field images: a large-scale vortex wake and a horizontal wake. The metrics—PSNR, SSIM, FID, and MAE—achieved values of 32.5831, 0.8612, 17.8514, and 0.0191, respectively, representing improvements of 4.54%, 2.82%, 5.18%, and 11.16% over the baseline DeOldify algorithm.